AI coding tools pricing, verified
We track pricing for AI coding tools straight from each vendor's official pricing page, not from press releases or secondhand roundups. Vendors in this space adjust prices and usage limits often, and a number quoted in a months-old blog post has a good chance of being wrong by the time you read it. Every entry in the table below links to a per-tool page with the full plan breakdown, key usage limits, and a dated history of every price change we've caught, each one sourced to where we found it. The Verified column tells you when we last confirmed a tool's numbers against the vendor's live page.
Cheapest paid plan · high → low
Recently changed from our log
- billingKimi Code — Kimi K3 became available in GitHub Copilot (Pro, Pro+, Max, Business, Enterprise) under usage-based billing at Moonshot list rates, hosted by GitHub on Fireworks AI rather than proxied to Moonshot; off by default on Business and Enterprise until an admin enables the policyAug 6, 2026
- billingCodex CLI — Correction on our side, not a vendor price change: OpenAI does publish the Pro 20x price at $200/mo, in the ChatGPT Voice section of the pricing page ("Pro 20x ($200/month): Unlimited voice access"). We had recorded it as unpublished because the plan card itself only says "From $100/month".Jul 31, 2026
- billingCursor — Cursor Start launched, a lower-priced plan for developers in India at ₹649/mo tax inclusive. It covers the Cursor Models pool (Grok 4.5 and Composer 2.5, both non-fast, Grok fixed at medium effort) plus Cloud Agents, but excludes the Other Models pool, on-demand usage, Bugbot, Auto, Automations, and the SDK. Not listed in the plan table above because it is India-only and priced in INR.Jul 28, 2026
- billingCursor — Auto mode is no longer unlimited (observed on this date; Cursor does not date the change). Auto now bills per token — the cost-saver setting at a flat $1.25 in / $6 out per 1M tokens, the balance/intelligence settings at the selected model's API rate. Plans now carry two pools: a generous included pool for Cursor's own models (Composer 2.5, Grok 4.5) plus the $20/$70/$400 "Other Models" third-party pool.Jul 28, 2026
Filter the ranking by billing model:
Pricing comparison
| Tool | Free tier | Entry paid | Power tier | Team | Verified |
|---|---|---|---|---|---|
| Aider | Yes | — | — | — | |
| Amazon Q Developer | Yes | $19/user/mo | $19/user/mo | — | |
| Antigravity | Yes | $4.99/mo | $199.99/mo | — | |
| Augment Code | No | $100/mo | — | $100/mo | |
| Claude Code | Yes | $20/mo | $200/mo | $25/user/mo | |
| Cline | Yes | — | — | — | |
| Codex CLI | Yes | $8/mo | $100/mo | $25/user/mo | |
| Cursor | Yes | $20/mo | $200/mo | $40/user/mo | |
| Gemini Code Assist | No | $19/user/mo | $54/user/mo | $19/user/mo | |
| GitHub Copilot | Yes | $10/user/mo | $100/user/mo | $19/user/mo | |
| JetBrains AI | Yes | $8.33/user/mo | $25/user/mo | $60/user/mo | |
| Kimi Code | Yes | $19/mo | $199/mo | — | |
| Windsurf (now Devin Desktop) | Yes | $20/mo | $200/mo | $80/mo |
How we verify
Primary sources only. For us that means the vendor's own pricing page, plus official blog posts and changelogs when a change is announced before the page updates. We never take a number from a third-party article, a comparison site, or a cached copy. If we can't confirm a figure on a page the vendor controls, it doesn't go in the table.
Some pricing pages make this harder than it should be. A few render prices with JavaScript, so a plain fetch returns nothing useful; others show different figures depending on your region. For those we check manually in a real browser from a US connection and keep a screenshot as backup, so a later dispute about what the page said on a given date has evidence behind it.
Any change to a price, to included usage, or to plan structure (a plan added, removed, or renamed) gets a changelog entry on that tool's page. Each entry carries the date we confirmed the change and a link to the source where we confirmed it. Silent limit cuts count too; those are the changes vendors are least likely to announce and the ones most likely to bite you mid-billing-cycle.
On schedule: we re-check the five most-used tools weekly and the rest biweekly. When a tool's verified date is more than 21 days old, the table above flags it with a "re-verification pending" badge. That badge doesn't mean the numbers are wrong, only that we haven't confirmed them recently enough to vouch for them, so treat that row with more caution until the date refreshes.